MLVU

MLVU is a dataset that contains 200 videos and 800 generated summaries, used to evaluate video-to-text summarization through multimodal question answering.

Papers using MLVU (13)

QEVA: A Reference-free Evaluation Metric For Narrative Video Summarization With Multimodal Question Answering2026Event-Anchored Frame Selection for Effective Long-Video Understanding2026Question-guided Visual Compression with Memory Feedback for Long-Term Video Understanding2026Adaptive Greedy Frame Selection for Long Video Understanding2026ForestPrune: High-ratio Visual Token Compression for Video Multimodal Large Language Models via Spatial-Temporal Forest Modeling2026ReMoRa: Multimodal Large Language Model based on Refined Motion Representation for Long-Video Understanding2026MSJoE: Jointly Evolving MLLM and Sampler for Efficient Long-Form Video Understanding2026Think-Clip-Sample: Slow-Fast Frame Selection for Video Understanding2026LiViBench: An Omnimodal Benchmark for Interactive Livestream Video Understanding2026Video-o3: Native Interleaved Clue Seeking for Long Video Multi-Hop Reasoning2026Towards Effective Long Video Understanding of Multimodal Large Language Models via One-shot Clip Retrieval2025Q-Frame: Query-aware Frame Selection and Multi-Resolution Adaptation for Video-LLMs2025FALCONEye: Finding Answers and Localizing Content in ONE-hour-long videos with multi-modal LLMs2025